CxO · Daily · April 9, 2026

the labor‑for‑silicon trade

In Q1 2026, nearly 80,000 tech employees were laid off.[2] Roughly half of those cuts were explicitly attributed to AI.[3] Oracle cut about 30,000 roles as part of a shift into AI data centers.[4]

Amazon shed 16,000 employees and Microsoft roughly 15,000, both explicitly to redirect spend into AI data center capacity.[5]

At the application layer, Salesforce removed 4,000 support roles using AI agents but later admitted those agents had serious reliability problems, reinforcing internal criticism that 'AI layoffs' are outrunning what current systems can reliably do.[6]

the compute stack stops being one company

NVIDIA is still buying speed, signing a $20 billion licensing deal for Groq’s IP and team and pushing its Groq 3 LPU architecture with Tensor‑First compute and ultra‑high SRAM bandwidth.[7]

At the same time, major customers like Uber and Anthropic are choosing Amazon’s custom silicon over Nvidia GPUs for some AI workloads, signaling real demand for non‑Nvidia accelerators.[8]

In China, Alibaba has stood up a data center with 10,000 of its own AI chips, further deconcentrating the global compute supply away from US GPU vendors.[9]

Qwen 3.5‑27B is reported to be 25 times cheaper than rival models targeting similar tasks.[10] Open‑weights systems like GLM‑5.1 and new open video models such as HappyHorse 1.0 are closing capability gaps with proprietary stacks, giving enterprises and developers credible non‑hyperscaler options.[11]

gated AI security and a $21B attack surface

Americans lost an estimated $21 billion to cybercrime last year.[12] Government‑official impersonation scams roughly doubled in 2025, triggering the FBI and DOJ’s Operation Masquerade crackdown.[13]

Nation‑state and criminal actors are already hitting physical infrastructure, from Iranian operations against US water and energy facilities to a ransomware attack that disrupted care at Signature Healthcare’s Brockton Hospital.[14]

Into that environment, Anthropic’s Claude Mythos is surfacing critical vulnerabilities across all major operating systems and browsers and scored 93.9% on the SWE‑bench Verified benchmark.[15]

Mythos is restricted to billion‑dollar companies, governments, and select researchers, even as Anthropic puts $100 million into Project Glasswing and Apple quietly uses the model to harden its own platforms.[16]

china’s parallel AI economy

Alibaba has opened a data center equipped with 10,000 of its own AI chips, rather than relying on US vendors.[17] Its Qwen 3.6 Plus model shows a five‑point reasoning gain over Qwen 3.5‑397B and performs on par with MiniMax‑M2.7, though the weights remain closed.[18]

Qwen is now widely recommended for AI applications inside China, and some users argue Qwen and related models like HappyHorse outperform many Western alternatives.[19]

Alibaba is already commercializing this stack, shipping the OpenClaw system and deploying the Wukong model to about 20 million DingTalk users.[20] The ATOM Report finds Chinese labs, especially Qwen, dominating open‑source LLM contributions even as there is a visible trend toward closed‑source WAN models and uncertainty about future open releases such as Wan2.5 or Wan2.6.[21]

jurisdictions and platforms as product variables

Japan has deliberately relaxed privacy laws to become the easiest country for AI development, allowing broader data use while raising questions about overreach and local infrastructure readiness.[22]

Texas is handing out more than $1 billion per year in tax breaks to data centers even as residents question whether its stressed grid and drought‑hit water systems can support more energy‑intensive facilities.[23]

Local sentiment in Texas is turning hostile, with critics describing the incentives as corrupt giveaways to wealthy corporations at the expense of public infrastructure and the working class.[24]

On the vendor side, Microsoft unilaterally terminated the VeraCrypt developer’s account and locked the WireGuard VPN developer’s account, meaning Secure Boot will block VeraCrypt system‑drive encryption after June 2026 and intensifying concerns about corporate control over critical open‑source tooling.[25]

Geopolitically, the Pentagon is trying to break China’s rare‑earths monopoly while simultaneously blacklisting Anthropic, and OpenAI, Anthropic, and Google are coordinating to block Chinese scraping of their models, underscoring how AI supply chains and access rights are now entangled with national security.[26]